Modelling of Electrodialysis with Bipolar Membranes processes using hybrid model supported by Artificial Neural Networks

Electrodialysis with bipolar membranes appears to be a promising process for thesimultaneous production of acid and alkaline solutions. This process can be integratedinto circular economy approaches for waste streams valorisation or utilised alone,especially in remote areas, to reduce transportation, storage and handling of thesehazardous chemicals. However, there is a lack of information on large-scale units in realoperational environments, and, as a result, there are no validated modelling tools that canbe used for its design, optimisation, simulation and control. The aim of the present PhDthesis is to design and test a semi-industrial electrodialysis with bipolar membranes unitand develop versatile modelling tools that can be adopted for the above-mentionedapplication. The experimental investigation focused on evaluating well established andnew process configurations and operational schemes, as well as testing the process insidean integrated treatment chain to valorise a seawater brine. The collected data were utilisedto develop the modelling tools. Firstly, a model with a first principles approach wasobtained and validated to simulate large-scale units also with complex stack configuration(i.e., internal staging). The model was subsequently modified to account also fornonstationary operations. In addition, the possibility of adopting innovative modellingtools was considered. For the first time, artificial neural network models were used tosimulate the Electrodialysis with bipolar membranes process. Finally, the combination offirst principles and data-driven models was considered to develop innovative hybridmodels with superior performance compared to the two types of models used alone. Theobtained results can guide the selection of the most appropriate process configurationdepending on the applications, while the proposed models can be selected as realisabletools to predict the process behaviour depending on the application.